A setup that turns a topic you give into a Turkish-voiced video and your own product's screen recording into a demo video

How the system fits together

Topic-to-video: topic and sources → OpenMontage script and scene plan → stock or generation service → Google Turkish voice → captions → local render → MP4. Product demo: recording with Recordly → scene selection → titles and audio → MP4.

• O
  OpenMontage required

  Open-source system that manages research, script, scene plan, voice and render in the topic-to-video path

  Tool guide

• 
  Claude Code required

  Coding assistant that opens the OpenMontage folder and runs the production, asking you step by step

  Tool guide

• 
  OpenAI Codex CLI optional

  An assistant that does the same job

  Alternative: If you use Codex rather than Claude Code

  Tool guide

• F
  FFmpeg required

  Local tool needed to combine audio, captions and video

  Official site

• 
  Google Cloud Text-to-Speech required

  Turkish voiceover (tr-TR Chirp 3 HD voices)

  About the tool

• P
  Pexels optional

  Source of licensed stock images and video

  Alternative: If you want visuals from stock rather than from a generation service

  Official site

• 
  Remotion optional

  OpenMontage's render engine that comes with the repository; you don't install it separately

  Alternative: When OpenMontage chooses this render path for the script; a license may be required depending on company size and commercial use

  Tool guide

• 
  HyperFrames optional

  A second render path OpenMontage can choose for videos heavy on HTML and motion graphics

  Alternative: If the script relies heavily on animated text and graphics and OpenMontage suggests this path

  Tool guide

• 
  Recordly required

  In the product demo path, records the screen and exports an MP4 with cropping, zoom, titles and audio added

  Tool guide

• 
  Raylight optional

  A tool for designing animated product promos from screenshots; the site also lists a separate 'Capture for Mac' app that records the Mac screen (this guide focuses on Windows)

  Alternative: If you want a designed animated promo scene rather than a real recording

  Tool guide

• S
  Shotbase optional

  Screen recording and automatic zoom on Mac

  Alternative: If you use only a Mac and accept the paid version

  Tool guide

• 
  AutoAE optional

  Ready-made motion graphics clip service; its site lists no screen recording feature. The free plan is for non-commercial use

  Alternative: If you're only looking for animated text or graphic clips and will consider the paid plan for a business video; the site doesn't list publisher information, so use it at your own judgment

  Tool guide

Before you start

Step-by-step setup

The steps you check stay only in this browser; they do not confirm that the setup works. You finished 0/7 steps.

1. Step 1 · In the terminal

   Check that the required programs are installed

   I finished this step

   What for: Keep the setup from stopping halfway in the next steps.

   1. Open Windows PowerShell. First check that Python, Git and Node.js are installed (Python 3.10 or later and Node.js 18 or later are required; 22 or later if you choose the HyperFrames path).
   2. If FFmpeg isn't installed, install it with the winget line below; after installing, close and reopen PowerShell.
   3. Check that the two check lines below print version information.
   WindowsFolder: Any folder

   › winget install --id Gyan.FFmpeg --version 9.0.2 --exactCopy

   Version pinned (as of October 17, 2026). Check the vendor documentation before moving to a newer version.

   Windows · macOS · LinuxFolder: Any folder

   › ffmpeg -versionCopy

   Windows · macOS · LinuxFolder: Any folder

   › node --versionCopy

   How you know it worked

   • ffmpeg -version prints a version number.
   • node --version prints version 18 or later.
   If you get stuck: problems specific to this step

   • The winget command isn't recognized
     Install FFmpeg from the Windows build link on the ffmpeg.org download page and add the folder containing ffmpeg.exe to the system path; then check with ffmpeg -version.
   • You use macOS or Linux
     The OpenMontage README gives brew install ffmpeg for macOS and sudo apt install ffmpeg for Debian-based Linux; the commands in this guide are for Windows PowerShell.
   Based on: OpenMontage README (pinned commit), winget FFmpeg package manifest (Gyan.FFmpeg 9.0.2), OpenMontage provider guide (docs/PROVIDERS.md)Get help with this step →
2. Step 2 · In the terminal

   Download and install OpenMontage at a specific version

   I finished this step

   What for: Get the repository at a version that won't change and install the dependencies.

   1. Create an empty folder named 'video-calismasi' under Documents and start PowerShell there.
   2. Download the repository and pin it to the specified commit. The repository has no published release tags, so it is pinned by commit hash.
   3. Create a virtual environment, activate it and install the Python dependencies. The requirements file sets lower bounds (for example >=), so second-level packages are not fully pinned.
   4. In the remotion-composer folder, install the Node packages at the versions in the lock file, then copy the settings file.
   5. Don't install the offline Piper voice: the Turkish Piper voice carries a data license that doesn't allow commercial use (see the voice step). The Makefile says that if Piper installation fails, voice falls back to cloud providers; the documentation doesn't separately confirm that a setup without Piper leaves the rest of the workflow intact. You'll check that the Google voice is ready at the end of the next step, where you add the key.
   WindowsFolder: Documents\video-calismasi

   › git clone https://github.com/calesthio/OpenMontage.git; git -C OpenMontage checkout c8066452325cb57c8fc1ba8c2a71b6e8842e6b65Copy

   Version pinned (as of October 17, 2026). Check the vendor documentation before moving to a newer version.

   WindowsFolder: Documents\video-calismasi

   › git -C OpenMontage log -1 --onelineCopy

   WindowsFolder: Documents\video-calismasi\OpenMontage

   › py -3 -m venv .venvCopy

   WindowsFolder: Documents\video-calismasi\OpenMontage

   › .\.venv\Scripts\Activate.ps1Copy

   WindowsFolder: Documents\video-calismasi\OpenMontage

   › python -m pip install -r requirements.txt # the file in OpenMontage commit c8066452325cb57c8fc1ba8c2a71b6e8842e6b65; the file itself uses >= ranges, only the commit is pinnedCopy

   Version pinned (as of October 17, 2026). Check the vendor documentation before moving to a newer version.

   WindowsFolder: Documents\video-calismasi\OpenMontage

   › cd remotion-composer; npm ci; cd ..Copy

   WindowsFolder: Documents\video-calismasi\OpenMontage

   › Copy-Item .env.example .envCopy

   How you know it worked

   • The output of git -C OpenMontage log -1 --oneline is a single line starting with c806645.
   • The folder shows .venv, remotion-composer/node_modules and the .env file.
   If you get stuck: problems specific to this step

   • The npm command throws an ERR_INVALID_ARG_TYPE error
     The OpenMontage README gives npx --yes npm install for this error on Windows; use it only if this error appears, because it doesn't apply the lock file versions exactly.
   • Activate.ps1 can't be run (script execution disabled)
     The Windows PowerShell execution policy may be blocking activation; ask for support at this step before changing the policy, and don't loosen general security settings at random.
   Based on: OpenMontage README (pinned commit), OpenMontage repository commit (AGPL-3.0, no release tag), npm ci docs, OpenMontage MakefileGet help with this step →
3. Step 3 · In the browser

   Add the keys for the Turkish voice and stock media to the .env file

   I finished this step

   What for: Give OpenMontage access to the Google Turkish voice and the stock sources.

   !You check this step yourself; the system should not move past it without your approval.

   1. Create a new project in your Google Cloud account, enable billing and turn on the Text-to-Speech API. Restrict the key so it only allows this API.
   2. .env.example has no GOOGLE_TTS_API_KEY line; add it as a new line at the end of .env and put the key there. The same file also has a GOOGLE_API_KEY line, and the same Google tool reads it too; fill in only one, don't mix them up.
   3. Don't put the key in a chat, an email, a screenshot or Git.
   4. If you like, open Pexels and Pixabay accounts and put their keys in the PEXELS_API_KEY and PIXABAY_API_KEY lines in .env; these lines already exist in the file, don't add them again. Both are optional. The Pixabay API documentation requires you to credit the source when showing results and sets a limit of 100 requests/60 sec; it also expects you to host images on your own server and cache them for 24 hours.
   5. When using the Pexels API, you're expected to show a visible link to Pexels and credit the photographer where possible; the default limit is 200 requests per hour and 20,000 per month. The free Google allowance is monthly; set up a budget alert in Google Cloud.
   .env (single line to add at the end) · iniCopy

   GOOGLE_TTS_API_KEY=

   The value is left empty as an example; write your own key after the equals sign and don't add the file to Git. The PEXELS_API_KEY and PIXABAY_API_KEY lines already come from .env.example.

   How you know it worked

   • The Text-to-Speech API shows as 'enabled' in Google Cloud.
   • The GOOGLE_TTS_API_KEY line in .env has a value and isn't stored anywhere else.
   • When you tell OpenMontage 'show the provider menu', Google TTS appears on the 'available' side of the list the assistant shows (if it stays on the unavailable side, the key or API setting is missing).
   If you get stuck: problems specific to this step

   • You get a permission or API-not-enabled error while generating voice
     OpenMontage's voice tool requires the Text-to-Speech API to be enabled for the key; enable the API in Google Cloud and check that the key restrictions allow Text-to-Speech.
   • You accidentally shared the key or added it to Git
     Delete that key in Google Cloud, create a new one and update .env with the new key; just deleting it from the file isn't enough.
   • Google TTS stays on the 'unavailable' side of the provider list
     Check that the key is on the right line in .env, that you saved the file and that you restarted the assistant; then ask it to show the menu again. If that doesn't fix it, ask for support at this step.
   Based on: OpenMontage Google TTS tool (google_tts.py), Google Cloud Text-to-Speech pricing, OpenMontage provider guide (docs/PROVIDERS.md), OpenMontage .env.example, Pexels API docs, Pixabay API docs, OpenMontage tool registry (tool_registry.py)Get help with this step →
4. Step 4 · In the terminal

   Install the word-timing package for captions

   I finished this step

   What for: Prepare the transcription component needed so captions appear in sync with the voice.

   1. OpenMontage's transcription tool requires the faster-whisper package; this package isn't in requirements.txt, and the Google voice tool doesn't produce timestamps either (the code has no such output). Install the package in CPU mode with the command below.
   2. Make sure the virtual environment is active ((.venv) appears at the start of the line).
   3. Tell the assistant explicitly that the caption language is Turkish. OpenMontage's transcription tool uses the 'base' model by default and auto-detects the language if none is given; this can produce incorrect text in Turkish. If the text is wrong, you can ask for a larger model (download size and time increase).
   4. The assistant may download an extra file (model) on the first transcription; before approving the download, ask what it will download. If the caption component doesn't work, word-timed captions may not appear; you check this by watching the video.
   Windows · macOS · LinuxFolder: Documents\video-calismasi\OpenMontage

   › python -m pip install faster-whisper==1.2.1Copy

   Version pinned (as of October 17, 2026). Check the vendor documentation before moving to a newer version.

   How you know it worked

   • The command finishes without errors, and python -m pip show faster-whisper shows version 1.2.1.
   If you get stuck: problems specific to this step

   • The command throws a build or CTranslate2 error
     Show the error text to the assistant and ask for support; Python should be 3.10 or later, as OpenMontage requires. Without this package you may not get word-timed captions.
   • Captions come out out of sync with the voice
     Have the assistant fix the caption text and timing; don't share the video until you're sure the captions are correct.
   Based on: OpenMontage transcription tool (transcriber.py), faster-whisper (PyPI), OpenMontage requirements.txtGet help with this step →
5. Step 5 · In a coding agent

   Give the assistant the topic, the sources and the target video

   I finished this step

   What for: Make sure the script is based on the sources you provide and reach the first approval checkpoint.

   !You check this step yourself; the system should not move past it without your approval.

   1. Open the OpenMontage folder in the coding assistant. The assistant reads the instruction files in the repository.
   2. Provide the following as input: the topic sentence, the target audience, the language (Turkish), vertical format (ask for the 9:16 profile; the aspect ratio is set in the proposal), a voiceover of about 45-60 seconds (this is an acceptance example, not a duration promise), source links, the brand color and the logo file.
   3. Ask for a Turkish Google voice (example name format: tr-TR-Chirp3-HD-Achernar) and tell it to take footage from licensed stock.
   4. Read the assistant's research summary, the proposal and the script. If there is an unsourced claim, a made-up figure or a brand that isn't yours, delete it and have it corrected; don't move on to the next stage without giving approval.

   How you know it worked

   • A research summary and a script document are created in the project folder.
   • The assistant waits for your approval of the script and scene plan.
   If you get stuck: problems specific to this step

   • The assistant asks which provider to use or doesn't show the voice options
     Ask the assistant to run OpenMontage's provider menu (provider_menu) and show which tools are ready; the README gives this command for the assistant.
   • The script contains figures with no source
     Remove those sentences or have them written with their source; OpenMontage's research stage is set up to cite sources, but the final check is yours.
   Based on: OpenMontage README (pinned commit), Google Cloud Text-to-Speech: voices and typesGet help with this step →
6. Step 6 · In a coding agent

   Approve the scenes, the Turkish voice and the captions by listening

   I finished this step

   What for: Keep mispronunciations, wrong captions or unlicensed footage out of the render.

   !You check this step yourself; the system should not move past it without your approval.

   1. Review the scene plan and the generated visuals scene by scene; OpenMontage puts a separate approval gate on visuals. Note the source of every stock clip (Pexels or Pixabay).
   2. Listen to the Turkish voice from start to finish: are proper names, numbers, the Turkish-specific letters and the stress correct?
   3. Read the captions alongside the voice: are the word timing, Turkish characters and line breaks right?
   4. Don't leave a scene that doesn't show your real product labeled as 'this is how our product works'; either replace it with a real screen recording (the other path) or write 'illustration' in the caption.

   How you know it worked

   • You gave the assistant your approval for the scenes, voice and captions in writing.
   • If there was a mispronunciation or caption error, you listened to the corrected version again.
   If you get stuck: problems specific to this step

   • The voice sounds robotic or reads incorrectly
     Have it choose another tr-TR voice from the same Google provider and rewrite the text around hard-to-pronounce words; every regeneration counts against your usage allowance.
   • The offline Piper Turkish voice was suggested
     The model card of Piper's tr_TR dfki voice lists the CC BY-NC-SA 4.0 data license (non-commercial use); don't use this voice for a business video.
   Based on: OpenMontage README (pinned commit), Pexels license, Pixabay content license summary, Google Cloud Text-to-Speech: voices and typesGet help with this step →
7. Step 7 · In a coding agent

   Render on your computer and check the MP4

   I finished this step

   What for: Turn the plan you approved into a single MP4 locally that can be reproduced.

   !You check this step yourself; the system should not move past it without your approval.

   1. When you approve the render, the assistant runs it on your computer with local Node and FFmpeg; no separate cloud render service is needed.
   2. The output file is written to the renders folder under the project folder, named final.mp4. Watch the video from start to finish: you check that voice and captions match and that there are no black frames or logo errors.
   3. To reproduce the same video, keep the project folder and the recorded approvals; ask the assistant to change a scene or the voice in the same project and render again.
   4. This guide doesn't publish the video. If you want to publish, use a separate publishing guide and give the approval yourself.

   How you know it worked

   • The file projects/<project-name>/renders/final.mp4 is created and plays.
   • You check the file's resolution and aspect ratio in the player or in the ffprobe output: ask for the vertical profile, but the actual ratio depends on the value in the assistant's proposal.
   If you get stuck: problems specific to this step

   • The HyperFrames render path was chosen and the command is downloading something
     OpenMontage calls HyperFrames with npx hyperframes and doesn't pin the version; the npm registry currently lists 0.8.167. If you want a pinned version, don't let it choose this path, or ask for support.
   • FFmpeg not found error during render
     Reopen PowerShell and check that the ffmpeg -version command works; if it doesn't, the FFmpeg folder is not on the system path.
   Based on: OpenMontage provider guide (docs/PROVIDERS.md), OpenMontage README (pinned commit)Get help with this step →

What you pay for

The items are separate, not a total; which ones you need depends on the path you set up. Prices change, so check the source of each item.

• SubscriptionCoding assistant

  OpenMontage needs an assistant that can read and run code; the assistant's own plan and price are on the developer's page. The documentation lists no charge for the OpenMontage repository itself. source

• API usageGoogle Text-to-Speech (Turkish Chirp 3 HD voice)

  According to Google's pricing page, the first 1 million characters each month are a free allowance, then $30 per million characters; billing must be enabled, and you are charged automatically if you exceed the allowance. (October 17, 2026) source

• API usageGeneration service (optional: AI video and images)

  If a generation service is chosen instead of stock, it is billed by usage. OpenMontage shows an estimate before spending; the default total cap is $10, the approval threshold for a single operation is $0.50, and both are adjustable. (October 17, 2026) source

• No costLicensed stock (Pexels, Pixabay)

  According to the Pexels license page, use is free and attribution is not required; selling on another stock platform, implying brand endorsement and portraying people negatively are prohibited. When using the Pexels API, however, a visible link to Pexels and, where possible, photographer credit are expected; the default limit is 200 requests per hour and 20,000 per month. For Pixabay, the full license, not the summary, is binding; selling content on its own and misleading use are prohibited. Keep the source of every file. The Pixabay API documentation also requires the source to be credited when results are shown and sets a limit of 100 requests/60 sec. source

• SubscriptionRemotion license

  Under the Remotion 4.x license, individuals, for-profit companies with up to 3 employees, non-profit organizations and evaluators not yet using it commercially can use it for free; for-profit organizations of a certain size must get a Company License. LICENSE.md says the license will change in Remotion 5.0. It isn't free for every business; if you're near the threshold, see remotion.pro/license. source

• No costOpenMontage and Recordly licenses

  Both are AGPL-3.0 open source. The documentation lists no charge for using them; if you'll modify and distribute their code, read the license terms. Third-party models and media are not under the same license. source

• Setup serviceSetup service

  You can do the setup yourself; if you'd like support from DijiKoçum on the steps where you get stuck, the scope is agreed in a call. source

• SubscriptionAutoAE (alternative)

  The free plan is listed on the site as 'Non-commercial use only', watermarked, 720p and limited to a one-time 500 credits; for a business video, look at the paid plans. source

Sources

• OpenMontage README (pinned commit) · Source code · read October 17, 2026
• OpenMontage provider guide (docs/PROVIDERS.md) · Source code · read October 17, 2026
• OpenMontage repository commit (AGPL-3.0, no release tag) · Source code · read October 17, 2026
• OpenMontage Google TTS tool (google_tts.py) · Source code · read October 17, 2026
• npm ci docs · Official documentation · read October 17, 2026
• winget FFmpeg package manifest (Gyan.FFmpeg 9.0.2) · Source code · read October 17, 2026
• FFmpeg download page · Vendor page · read October 17, 2026
• Google Cloud Text-to-Speech: voices and types · Official documentation · read October 17, 2026
• Google Cloud Text-to-Speech pricing · Official documentation · read October 17, 2026
• Remotion license (LICENSE.md) · Source code · read October 17, 2026
• HyperFrames (HeyGen) repository and npm registry entry · Source code · read October 17, 2026
• Recordly README · Source code · read October 17, 2026
• Recordly v1.4.0 release and SHA256SUMS.txt · Source code · read October 17, 2026
• Pexels license · Vendor page · read October 17, 2026
• Pixabay content license summary · Vendor page · read October 17, 2026
• Piper tr_TR dfki voice model card · Source code · read October 17, 2026
• Shotbase · Vendor page · read October 17, 2026
• Raylight · Vendor page · read October 17, 2026
• AutoAE · Vendor page · read October 17, 2026
• OpenMontage transcription tool (transcriber.py) · Source code · read October 17, 2026
• OpenMontage .env.example · Source code · read October 17, 2026
• OpenMontage Makefile · Source code · read October 17, 2026
• faster-whisper (PyPI) · Vendor page · read October 17, 2026
• Pexels API docs · Vendor page · read October 17, 2026
• AutoAE pricing · Vendor page · read October 17, 2026
• Recordly language configuration (src/i18n/config.ts) · Source code · read October 17, 2026
• Pixabay API docs · Vendor page · read October 17, 2026
• OpenMontage tool registry (tool_registry.py) · Source code · read October 17, 2026
• OpenMontage requirements.txt · Source code · read October 17, 2026